entropicdrift

Time is not a flat circle, but an endless spiral. We’ve come back around to Dogpile

I’ve been using the same 10+ year old 128G SSD for transcoding, cache, and metadata storage for like 6 years now across multiple tiny servers with limited RAM. My current one has 16G of RAM, which is more than enough for JF, but not enough to store a whole movie transcode, let alone multiple at once. Bad small old SSDs are a lot cheaper and easier to come by than RAM these days, that’s all I’m saying. No hate of course, your setup makes sense for what you have, I’m just saying my way is cheaper and more than fast enough for people just building their servers now.

tmpfs is overkill. I just use a crappy old 128G SSD and it’s still way faster than needed for all of my transcodes combined.

Anyway just my 2 cents.

Don’t trust AI to know what they’re doing for you. The only time they work reliably as a tool is when you already know what you’re doing enough to spot their errors.

AI is the wrong tool here. You need to do real internet research.

A new version was just released. One of the new features is parallel federation, in addition to reduced CPU usage. Hopefully if we upgrade to it, we’ll see reduced federation lag with lemmy.world

@SDF@lemmy.sdf.org: Lemmy is now up to 0.18.4. This update fixes collapsed comments, and apparently 0.18.3 should have a massive performance improvement with regards to database accesses, so I think it would be good to update.

If you scroll to the bottom of the page it shows the version on the bottom-right. Currently says BE: 0.18.2 The BE stands for Back-End. Sometimes the Front-End of Lemmy gets updated separately from the Back-End, hence the distinction.

I’m keeping an eye on this PR

Once that one’s merged it should massively improve database performance under load. Instead of doing tons of full table reads with joins to the communities table, it’ll be doing much smaller, faster reads. Currently those full reads happen for every up/down vote.

Converting those blind full-table scans with joins into smaller targeted reads is gonna be a performance gamechanger

Edit: it was just merged!